Aider Polyglot
Harbor nativeAider Polyglot: 6种语言diff编辑
Verified71%Coding
- Samples
- 225
- Released
- 2024
- Scores
- 61
GeoNatureAgent (GNA)
GeoNatureAgent (GNA): A Framework and Benchmark for Pre-Production Evaluation of Tool-Using Agents on Geospatial and Environmental Tasks
A 103-task benchmark evaluating tool-using LLM agents on geospatial/environmental tasks via a fixed MCP server interface and deterministic scorer.
How to run
ltb · geonatureagent-gna
$ npm i -g lettokenburn-cli$ ltb bench info geonatureagent-gna
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
9 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Sonnet 4 (2025-05-14) | 60.8 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| DeepSeek V3.2 | 56.3 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| GLM-5 | 50.2 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemini 2.5 Pro | 48.0 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| GPT-4o | 41.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Qwen3 235B | 41.2 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| GPT-oss 120B | 34.1 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| llama-4 Scout | 26.9 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemma 3 27B | 15.8 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
GNA features a 103-task suite (93 main + 10 expansion) across 18 categories, testing agent interaction with a self-hostable geospatial API serving environmental indicators.
It uses a deterministic eight-check scorer without LLM-as-judge. Claude Sonnet 4 achieved the highest capability at 61.7%, highlighting a gap between general GIS benchmarks and strict end-to-end evaluation.